WO2014174760A1 - Dispositif d'analyse d'action, procede d'analyse d'action et programme d'analyse d'action - Google Patents

Dispositif d'analyse d'action, procede d'analyse d'action et programme d'analyse d'action Download PDF

Info

Publication number
WO2014174760A1
WO2014174760A1 PCT/JP2014/001745 JP2014001745W WO2014174760A1 WO 2014174760 A1 WO2014174760 A1 WO 2014174760A1 JP 2014001745 W JP2014001745 W JP 2014001745W WO 2014174760 A1 WO2014174760 A1 WO 2014174760A1
Authority
WO
WIPO (PCT)
Prior art keywords
acoustic
information
event
time difference
analysis
Prior art date
Application number
PCT/JP2014/001745
Other languages
English (en)
Japanese (ja)
Inventor
亮磨 大網
博義 宮野
孝文 越仲
宝珠山　治
真宏 谷
Original Assignee
日本電気株式会社
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日本電気株式会社 filed Critical 日本電気株式会社
Priority to JP2015513507A priority Critical patent/JP6344383B2/ja
Priority to US14/786,931 priority patent/US9761248B2/en
Publication of WO2014174760A1 publication Critical patent/WO2014174760A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
    • G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
    • G10L25/57—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination for processing of video signals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/439—Processing of audio elementary streams
    • H04N21/4394—Processing of audio elementary streams involving operations for analysing the audio stream, e.g. detecting features or characteristics in audio streams
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
    • H04N7/188—Capturing isolated or intermittent images triggered by the occurrence of a predetermined event, e.g. an object reaching a predetermined position

Definitions

  • the present invention relates to a behavior analysis apparatus, a behavior analysis method, and a behavior analysis program that perform video surveillance using video information and audio information.
  • a technology to monitor people using acoustic information and video information For example, a specific sound pattern is detected from the sound signal, a surrounding image from which the sound signal is acquired is acquired, and processing such as enlargement, filtering, and interpolation is performed, or a surrounding solid from which the sound signal is acquired.
  • processing such as enlargement, filtering, and interpolation is performed, or a surrounding solid from which the sound signal is acquired
  • a method for facilitating specification of an abnormality by generating an image see, for example, Patent Document 1.
  • sound generated in the monitoring area and images of major locations are recorded using acoustic sensors and image sensors, and specific events (events) are detected by analysis of acoustic data and moved based on the detection results.
  • Patent Document 2 There is a method of tracking a body, acquiring image data of a moving body, and performing image analysis (see, for example, Patent Document 2). Both of the methods described in Patent Document 1 and Patent Document 2 are techniques for executing another image processing using voice or sound as a trigger.
  • crowd behavior is a set of individuals to be subjected to behavior analysis.
  • a method for analyzing the crowd behavior there is a method for determining whether it is a single group or an event (fight, crime, etc.) using acoustic analysis and image analysis (for example, Patent Document 3). reference.).
  • the acoustic event is an event extracted from voice or sound input by the microphone.
  • the acoustic event is, for example, an impact sound such as an explosion sound or a gunshot, or a human cry.
  • the video event is an event extracted from the video of the camera.
  • a video event is, for example, a crowd movement.
  • the present invention provides a behavior analysis apparatus, a behavior analysis method, and a behavior analysis program capable of analyzing a crowd behavior more accurately even when a time difference occurs between an acoustic event and a video event. For the purpose.
  • the behavior analysis apparatus analyzes the input acoustic information, generates an acoustic analysis information that represents the characteristics of the acoustic information, and the crowd after the acoustic event specified by the acoustic analysis information occurs.
  • the time difference determination unit for determining the time difference required until an event corresponding to the sound event occurs in the captured input image, the input image, the sound analysis information, and the time difference, the crowd behavior corresponding to the sound event is determined.
  • an action analysis unit for analysis.
  • the behavior analysis method analyzes input acoustic information, generates acoustic analysis information representing the characteristics of the acoustic information, and inputs an image in which a crowd is photographed after an acoustic event specified by the acoustic analysis information occurs. A time difference required until an event corresponding to the acoustic event occurs in the video is determined, and crowd behavior corresponding to the acoustic event is analyzed using the input video, the acoustic analysis information, and the time difference.
  • the behavior analysis program analyzes the input acoustic information in the computer, generates acoustic analysis information representing the characteristics of the acoustic information, and the crowd after the acoustic event specified by the acoustic analysis information occurs. Analyzing crowd behavior corresponding to an acoustic event using the input video, acoustic analysis information, and the time difference to determine the time difference required for the event corresponding to the acoustic event to occur in the input video captured It is characterized in that the processing is executed.
  • Embodiment 1 FIG. A first embodiment of the present invention will be described below with reference to the drawings.
  • FIG. 1 is a block diagram showing the configuration of the first embodiment of the behavior analysis apparatus according to the present invention.
  • the behavior analysis apparatus includes a voice / acoustic analysis unit 10, a time difference determination unit 20, and a crowd behavior analysis unit 30.
  • the voice / acoustic analysis unit 10 inputs voice / acoustic information (hereinafter simply referred to as acoustic information).
  • the acoustic information is information including human voice and sound transmitted from the surroundings.
  • the voice / acoustic analysis unit 10 inputs a voice / acoustic signal (hereinafter simply referred to as an acoustic signal) via a microphone. Any number of microphones may be connected to the behavior analysis device.
  • the voice / acoustic analysis unit 10 analyzes an acoustic signal.
  • the voice / acoustic analysis unit 10 is voice / acoustic analysis information (hereinafter simply referred to as acoustic analysis information) indicating the analysis result of the acoustic signal, for example, a feature quantity extracted from the acoustic signal (hereinafter referred to as acoustic feature quantity). Is generated.
  • the time difference determination unit 20 inputs acoustic analysis information from the voice / acoustic analysis unit 10.
  • the time difference determination unit 20 determines the time difference between the position of the microphone and the camera and the sound event and the video event, and generates time difference information indicating the time difference.
  • the crowd behavior analysis unit 30 inputs video information.
  • the crowd behavior analysis unit 30 inputs video information via a camera. Note that any number of cameras may be connected to the behavior analysis device.
  • the crowd behavior analysis unit 30 may input a plurality of pieces of video information from one camera.
  • the crowd behavior analysis unit 30 analyzes the crowd behavior based on the video information input from the camera, the time difference information input from the time difference determination unit 20, and the acoustic analysis information input from the voice / acoustic analysis unit 10.
  • An analysis result (hereinafter referred to as a crowd behavior determination result) is generated.
  • the crowd subject to the behavior analysis includes not only a single person but also a person who moves on a car, a motorcycle, a bicycle, or the like.
  • the crowd behavior analysis unit 30 includes a video crowd behavior analysis unit 31 and an analysis result integration unit 32 as shown in FIG.
  • FIG. 2 is a block diagram illustrating a configuration of the crowd behavior analysis unit according to the first embodiment.
  • the video crowd behavior analysis unit 31 analyzes the crowd behavior from the video taken by the camera, and generates video analysis information indicating the analysis result, for example, a feature quantity extracted from the video (hereinafter referred to as video feature quantity).
  • the analysis result integration unit 32 integrates the acoustic analysis information and the video analysis information in consideration of the time difference given by the time difference information, and determines crowd behavior based on the integrated result.
  • the analysis result integration unit 32 generates and outputs a crowd behavior determination result including the determination result.
  • the voice / acoustic analysis unit 10, the time difference determination unit 20, and the crowd behavior analysis unit 30 are realized by, for example, a computer that operates according to a behavior analysis program.
  • the CPU reads the behavior analysis program, and operates as the voice / acoustic analysis unit 10, the time difference determination unit 20, and the crowd behavior analysis unit 30 according to the program.
  • the voice / acoustic analysis unit 10, the time difference determination unit 20, and the crowd behavior analysis unit 30 may be realized by separate hardware.
  • the voice / acoustic analysis unit 10 analyzes an acoustic signal input from a microphone.
  • the voice / acoustic analysis unit 10 analyzes an acoustic feature, for example, a loudness and a kind of sound.
  • the voice / acoustic analysis unit 10 outputs acoustic analysis information including the analysis result to the time difference determination unit 20.
  • the time difference determination unit 20 determines a time lag (time difference) between the sound event specified from the sound feature amount indicated by the sound analysis information and the video event. That is, the time difference determination unit 20 determines the time difference between the abnormality detection time generated in the acoustic information and the abnormality detection time generated in the video information.
  • the time difference determination unit 20 determines the distance between the position where the acoustic event has occurred and the position where the video event is specified.
  • the “position where the acoustic event has occurred” is the position where the acoustic information is acquired, that is, the installation position of the microphone.
  • the “position where the video event is specified” is set as a shooting area of the camera, specifically, a position monitored by the camera.
  • the distance between the position where the acoustic event occurs and the position where the video event is specified is simply expressed as the distance between the microphone and the camera.
  • the time difference determination unit 20 After determining the distance between the microphone and the camera, the time difference determination unit 20 determines a time difference in anomaly that occurs between the acoustic information and the video information based on the distance, and generates time difference information indicating the time difference.
  • the time difference determination unit 20 holds time difference modeling information in advance.
  • the time difference modeling information is the time difference required from the occurrence of an acoustic event to the effect of the camera image (that is, the time difference required from the occurrence of the acoustic event to the occurrence of the video event) as a microphone. Information modeled according to the distance from the camera.
  • time difference modeling information an assumed time difference is set in advance from the distance between the microphone and the camera.
  • a time difference learned from an event that occurred in the past may be set in the time difference modeling information.
  • a time difference determined heuristically may be set in the time difference modeling information.
  • the time difference modeling information is, for example, a parameter of a calculation formula for obtaining a time difference according to a distance, or a look-up table describing a relationship between the distance and the time difference in a table.
  • the time difference may be modeled in consideration of not only the distance but also acoustic features such as the magnitude and frequency of the acoustic. For example, when modeling the time difference, the range in which the sound can be heard directly is estimated from the loudness and frequency of the explosion sound, etc., and if the distance between the microphone and the camera is within that range, the time difference is shortened. If it is farther than that, it may be modeled so as to increase the time difference. For example, when the sound input by the microphone is a sound that travels far away with a loud sound, such as an explosive sound, or when the sound includes a lot of high-frequency components and resonates far away, the time difference determination unit 20 uses the microphone.
  • the time difference determined from the distance between the camera and the camera is set to a shorter value.
  • the time difference modeling information is a parameter of a calculation formula (or a mathematical model) using distance and acoustic features as input variables.
  • the time difference determination unit 20 can obtain the time difference between the acoustic event and the video event more accurately based on the volume and type of sound indicated by the acoustic analysis information. That is, the time difference determination unit 20 can obtain a time difference in consideration of the volume and type of sound input by the microphone.
  • the time difference determination unit 20 may calculate the time difference as a distribution having a certain width instead of a single value. This is because a certain amount of variation can occur in the estimated value of the time difference. Specifically, for example, when the magnitude of the sound is large, the accuracy of estimation of the time difference is increased, so that the estimation width of the time difference is shortened. Also, when the acoustic level is small, the estimation accuracy tends to decrease and the estimation width of the time difference tends to increase. In addition, the accuracy of time difference estimation tends to decrease as the distance between the camera and the microphone increases. The time difference determination unit 20 outputs such a time difference as a distribution in consideration of such a tendency.
  • the time difference determination unit 20 sets the distribution of the time difference ⁇ to q ( ⁇ ), and generates information describing (representing) the distribution of q ( ⁇ ) as the time difference information. For example, when q ( ⁇ ) can be approximated by a normal distribution, the time difference determination unit 20 outputs an expected value (average value) of ⁇ and a variance value as time difference information.
  • the distribution shape of q ( ⁇ ) is not limited to a normal distribution, and a distribution such as a BPT (Brownian Timeline) distribution may be used for q ( ⁇ ).
  • the time difference determination unit 20 determines the time difference using both the distance between the microphone and the camera and the result of the acoustic analysis.
  • the time difference determination unit 20 outputs the time difference information to the analysis result integration unit 32 of the crowd behavior analysis unit 30.
  • the analysis result integration unit 32 inputs an acoustic feature amount used to analyze an abnormal state of sound from the voice / acoustic analysis unit 10. Further, the analysis result integration unit 32 inputs from the video crowd behavior analysis unit 31 video feature values used for analyzing the abnormal state of the crowd from the videos. The analysis result integration unit 32 integrates the acoustic feature quantity and the video feature quantity in consideration of the time difference indicated by the time difference information.
  • the probability that an event occurs at time t is Pa (t) as the probability obtained from the acoustic feature indicated by the acoustic feature amount.
  • the function G a is a function that estimates the probability that an event will occur from the value of each acoustic feature.
  • Function G a can be modeled by learning each acoustic features extracted from both the data of the detected subject to abnormal sound and other sounds. Alternatively, it may be determined a model of the function G a heuristically.
  • the probability that an event will occur at time t is Pv (t) as the probability obtained from the video feature indicated by the video feature amount.
  • the function Gv is a function that estimates the probability of occurrence of an event from the value of each video feature quantity.
  • Function G v can be modeled by learning each image feature extracted from both data of the video data and other video data of the abnormal state to be detected. Alternatively, it may be determined the model of function G v heuristically. In this case, the analysis result integration unit 32 calculates the probability P (t) that the event at time t is considered to have occurred by the following equation.
  • ⁇ represents the time difference indicated by the time difference information, that is, the time lag between the acoustic event and the video event.
  • the analysis result integration unit 32 integrates Pa (t) and Pv (t) in consideration of the time difference indicated by the time difference information.
  • the analysis result integration unit 32 calculates the value of P (t) directly from the values of the acoustic feature amount and the video feature amount without obtaining Pa (t) and Pv (t) using the following formula. May be.
  • the function G is a function for estimating the probability that an event will occur from the values of each acoustic feature quantity and each video feature quantity.
  • the probability that an event at the time t will occur is given by the following equation.
  • the analysis result integration unit 32 may directly calculate the value of P (t) using the following equation without obtaining Pa (t) and Pv (t).
  • the acoustic feature amount includes MFCC (Mel-Frequency Cepstrum Coefficients), FFT (Fast Fourier Transform), wavelet (Wavelet) transform coefficient, and the like extracted by transforming the acoustic signal.
  • MFCC Mel-Frequency Cepstrum Coefficients
  • FFT Fast Fourier Transform
  • Wavelet Wavelet transform coefficient
  • absolute value of volume amount of change of sound (secondary difference), direction of arrival of sound, speech recognition result of specific keyword (probability of recognition, frequency of recognition, specific keyword issued Number of speakers).
  • Video features include optical flows (information indicating apparent movement) in the video and histograms of directions and intensities obtained by summing them, or histograms obtained by summing and multiplexing them in various time widths, that is, There are histograms calculated at multiple time resolution, human detection results, and the like.
  • the analysis result integration unit 32 detects an abnormal state of the crowd from the result of integrating the acoustic feature quantity and the video feature quantity. For example, the analysis result integration unit 32 determines that an abnormal state of the crowd has occurred when the value of P (t) exceeds a preset threshold value.
  • the abnormal state refers to a state that is not steady.
  • the analysis result integration unit 32 outputs a crowd behavior determination result.
  • the analysis result integration unit 32 outputs the crowd action determination result only when a predetermined state to be detected, that is, an abnormal situation is detected, without being output during normal times.
  • the analysis result integration unit 32 may output a crowd behavior determination result indicating that it is normal, that is, that no abnormal state is detected.
  • the crowd behavior determination result includes, for example, a place where the abnormal behavior of the crowd is detected (for example, a position monitored by the camera), detection time, information indicating the type of abnormal behavior (for example, a predetermined event ID), A value indicating the degree of abnormal behavior and an event determination score (likelihood) indicating the likelihood of event determination are included.
  • information on the video when an abnormal behavior is detected and information on the area where the event was detected in the video (for example, where a specific person is running) It may be included in the determination result.
  • the analysis result integrating unit 32 further detects abnormalities such as explosions, fires, walk-throughs, and other violent incidents using the results of speech recognition and acoustic type determination. Reclassify behavior.
  • the analysis result integration unit 32 further reclassifies the abnormal behavior into theft or injury using the result of voice recognition.
  • the abnormal behavior shown in (3) above is an abnormal behavior that does not fall under (1) or (2), although some abnormal situation has been detected.
  • an index indicating the seriousness of the situation for example, a level is determined in advance as the degree of abnormal behavior.
  • the analysis result integration unit 32 determines the level based on the moving speed of the person, the number of persons escaping, or the loudness of the voice uttered at that time, and includes the determination result in the crowd action determination result.
  • a determination unit that outputs a level value from each input feature value value is generated in advance by regression learning or the like, and the analysis result integration unit 32 determines the level using the determination unit. Good.
  • the analysis result integration unit 32 may include additional information corresponding to the type of event in the crowd behavior determination result. For example, when the type of the abnormal behavior is (1), the analysis result integration unit 32 may include information on the movement direction of the crowd and the congestion status in the crowd behavior determination result. When the type of abnormal behavior is (2) above, the analysis result integrating unit 32 includes the personal characteristics (for example, characteristics of clothes, etc.), moving direction, and speed of the criminal candidate taking abnormal behavior in the crowd behavior determination result. You may make it let. Further, when the analysis result integration unit 32 deviates from the angle of view, the analysis result integration unit 32 may include the time and direction in the crowd action determination result.
  • the analysis result integration unit 32 may include the time and direction in the crowd action determination result.
  • the analysis result integration unit 32 may report the crowd behavior determination result to the security room as an alert.
  • the analysis result integration unit 32 may include an instruction for executing a predetermined operation in the crowd behavior determination result.
  • the analysis result integration unit 32 outputs the crowd action determination result to a device or the like that controls the emergency door, thereby opening the emergency exit. , You can sign the route to the emergency exit.
  • the analysis result integration unit 32 predicts a camera that is likely to show the next criminal candidate person from the moving direction, speed, time of departure from the angle of view, and the like. By outputting the crowd action determination result, the image of the camera can be output to the display device in the security room.
  • the analysis result integration unit 32 can control the direction of the camera, the zoom rate, and the like so that the face of the criminal candidate person can be easily captured.
  • the analysis result integration unit 32 performs collation based on person characteristics when a person who may be a criminal appears and determines that the person is highly likely to be the same person as the criminal.
  • the camera can keep tracking.
  • the analysis result integration unit 32 determines the abnormal state of the crowd from the result of integrating the acoustic analysis result and the video analysis result in consideration of the time difference. Therefore, even when the time when the abnormality is detected in the acoustic information is different from the time when the abnormality is detected in the video information, the abnormal behavior of the crowd can be reliably detected. Therefore, it becomes possible to analyze the behavior of the crowd more accurately.
  • the time difference determination unit 20 considers the positions of the microphone and the camera when determining the time difference between the audio event and the video event. For example, when the microphone for acoustic analysis and the camera for crowd behavior analysis are separated from each other to some extent, the time required for movement between them can be added to the time difference. Thereby, the abnormal behavior of the crowd can be accurately detected regardless of the installation position of the microphone and the camera.
  • the time difference determination unit 20 determines the time difference between the acoustic event and the video event, the loudness and type of sound input by the microphone are taken into consideration. Therefore, the time difference determined from the distance between the microphone and the camera can be updated to an optimum value based on the volume and type of sound.
  • the configuration of the second embodiment of the behavior analysis device is the same as that of the first embodiment.
  • the crowd behavior analysis unit 30 includes an event classification unit 33 in addition to the video crowd behavior analysis unit 31 and the analysis result integration unit 32.
  • FIG. 3 is a block diagram illustrating a configuration of the crowd behavior analysis unit according to the second embodiment.
  • the event classification unit 33 inputs acoustic analysis information.
  • the event classification unit 33 classifies the event based on the acoustic analysis information, and generates event classification information including the classification result.
  • the event classification information is used for controlling video crowd behavior analysis in the video crowd behavior analysis unit 31.
  • the video crowd behavior analysis unit 31 adjusts the parameters of the video crowd behavior analysis and switches the algorithm based on the event classification information generated by the event classification unit 33. At this time, the video crowd behavior analysis unit 31 performs adjustment of the parameter and switching of the algorithm in consideration of the time difference determined by the time difference determination unit 20.
  • the video crowd behavior analysis unit 31 needs to complete parameter adjustment and algorithm switching between the occurrence of an acoustic event and the occurrence of a video event. Therefore, for example, when the video crowd behavior analysis unit 31 determines that parameter adjustment or algorithm switching cannot be performed within the time difference determined by the time difference determination unit 20, parameter adjustment or algorithm switching is simplified. Or prevent it from being executed.
  • the video crowd behavior analysis unit 31 determines that the time required for the learning does not end within the above time difference. Avoid switching.
  • the video crowd behavior analysis unit 31 determines that the calculation resource is distributed among a plurality of cameras and the time required to actually change the calculation resource allocation is equal to or greater than the time difference, Do not change the allocation of calculation resources, or adjust the calculation resources more simply.
  • the video crowd behavior analysis unit 31 analyzes the crowd behavior in the video using the adjusted parameters and the switched algorithm.
  • the video crowd behavior analysis unit 31 outputs information obtained by the analysis as video analysis information.
  • the analysis result integration unit 32 integrates the acoustic analysis information and the video analysis information in consideration of the time difference, and determines crowd behavior based on the integrated result, as in the first embodiment.
  • the video crowd behavior analysis unit 31 determines that an abnormality has been detected from the acoustic information. That is, in this embodiment, the voice / acoustic analysis unit 10 outputs the acoustic analysis information to the time difference determination unit 20 when an abnormality is detected from the acoustic information. Then, the time difference determination unit 20 that has input the acoustic analysis information outputs the time difference information to the video crowd behavior analysis unit 31.
  • the voice / acoustic analysis unit 10 outputs the acoustic analysis information to the event classification unit 33 at the same time.
  • the event classification unit 33 classifies the event type of the crowd behavior, generates event classification information indicating the classification result, and outputs the event classification information to the video crowd behavior analysis unit 31.
  • the crowd behavior to be analyzed includes situations where a general person escapes in a group like terrorism (case A) or a specific person (such as snatching) There is a situation (referred to as case B) in which the criminal) runs away between the crowds.
  • the event classification unit 33 classifies these cases using acoustic information, and the video crowd behavior analysis unit 31 controls crowd behavior analysis by video based on the information.
  • the crowd behavior to be analyzed is not limited to Case A and Case B. There may be any number of types of crowd behavior cases to be analyzed.
  • the event classification unit 33 determines to be case A when it is found that a plurality of persons scream. In addition, for example, when it is found that one or a small number of persons are screaming, the event classification unit 33 determines as case B.
  • the voice / acoustic analysis unit 10 may perform voice recognition on the scream and the event classification unit 33 may analyze the utterance content to distinguish between the two.
  • the event classification unit 33 may determine that the case is A.
  • this determination does not need to be alternative, for example, the event classification
  • the video crowd behavior analysis unit 31 adjusts or switches video analysis parameters and algorithms based on, for example, event classification information in a crowd analysis using video. Specifically, the video crowd behavior analysis unit 31 adjusts or switches parameters and algorithms for analyzing the movement of the crowd.
  • the video crowd behavior analysis unit 31 determines, for example, a discrete motion in a group using a discriminator that detects a discrete motion pattern.
  • the video crowd behavior analysis unit 31 adjusts video analysis parameters so as to analyze the overall movement tendency. For example, the video crowd behavior analysis unit 31 reduces the spatial resolution of motion analysis so that the entire screen can be analyzed uniformly. Note that the video crowd behavior analysis unit 31 may gradually refine the motion analysis from a coarse resolution according to the calculation resource. Further, in a situation like Case A, an abnormal state is photographed by many surrounding cameras.
  • the behavior analysis device requires more computing resources than in normal times. Therefore, the video crowd behavior analysis unit 31 allocates the calculation resource allocated to the video analysis of each camera so that the calculation resource is not allocated to the processing of a specific camera and the video of another camera cannot be analyzed. adjust. For example, the video crowd behavior analysis unit 31 reduces the frame rate of analysis of each camera.
  • the video crowd behavior analysis unit 31 adjusts the video analysis parameters so that the movement of each individual person can be accurately followed. For example, the video crowd behavior analysis unit 31 increases the frame rate and the spatial resolution so that the reliability of the extracted motion is increased.
  • the video crowd behavior analysis unit 31 activates a module for performing processing for analyzing the detected part, and only the part is more detailed. You may make it analyze to.
  • the video crowd behavior analysis unit 31 sets motion analysis parameters for both cases based on the likelihood information. You may make it set to the value according to likelihood. For example, the video crowd behavior analysis unit 31 may control the temporal and spatial resolution of the motion analysis based on the likelihood information.
  • the video crowd behavior analysis unit 31 adjusts the spatial resolution (image size), density, and frame rate for calculating the optical flow according to the likelihood of case A and case B.
  • the density is a parameter used for determining whether the optical flow is obtained in units of pixels or every several pixels.
  • the video crowd behavior analysis unit 31 switches the algorithm used for calculating the optical flow according to the likelihood of the case A and the case B.
  • the video crowd behavior analysis unit 31 adjusts the video analysis parameters as follows.
  • the video crowd behavior analysis unit 31 determines parameters based on the following policy. (1) See the optical flow evenly throughout. (2) Do not increase temporal resolution and spatial resolution. (3) If the frame is overloaded, reduce the frame rate.
  • the video crowd behavior analysis unit 31 determines parameters based on the following policy. (1) Increase both temporal resolution and spatial resolution. (2) However, it is not necessary to view the optical flow as a whole, and it is only necessary to detect the movement of a person coming from a certain direction (that is, the direction in which the acoustic event is detected). Therefore, the temporal resolution and spatial resolution are increased so that the direction can be analyzed particularly finely.
  • the video crowd behavior analysis unit 31 determines the parameter value of each case in advance according to the above policy, and adjusts the parameter as follows according to the likelihood of case A and case B.
  • the video crowd behavior analysis unit 31 sets the spatial resolution to ⁇ 0 times the original image, and calculates the flow for each n 0 pixels.
  • the frame rate in the normal state is assumed to be f 0.
  • these parameters when determined as case A are represented by ⁇ A , n A , and f A
  • these parameters when determined as case B are represented as ⁇ B , n B , and f B.
  • the video crowd behavior analysis unit 31 sets each parameter to, for example, Calculate as follows.
  • the density may be biased according to the likelihood of case B.
  • the video crowd behavior analysis unit 31 switches the calculation algorithm according to the values of p A and p B.
  • a frame rate can be expressed as follows.
  • F (p A , p B ) is a function for calculating the frame rate.
  • F (p A , p B ) is not limited to a linear function, and various functions can be used.
  • the video crowd behavior analysis unit 31 may change the feature amount to be extracted according to the type of the case of the crowd behavior, for example, to extract a feature amount other than movement.
  • the video crowd behavior analysis unit 31 may change a dictionary (a motion pattern to be detected) included in the classifier.
  • the event classification unit 33 classifies the types of events of the crowd behavior, and the video crowd behavior analysis unit 31 determines the parameters of the video crowd behavior analysis and the like based on the classification result. Adjust or switch algorithms. Thereby, crowd behavior can be analyzed more accurately. Further, it is possible to efficiently use calculation resources. Further, even when there are a plurality of cameras, it is possible to appropriately distribute calculation resources allocated to video analysis of each camera. On the other hand, in the method described in Patent Document 3, since the acoustic analysis and the image analysis are performed independently, the parameters of the image analysis cannot be changed using the acoustic analysis result. For this reason, calculation resources may be consumed more than necessary. Further, in the method described in Patent Document 3, there is a possibility that a situation occurs in which calculation resources are allocated only to processing of a specific camera, and video of other cameras cannot be analyzed.
  • FIG. 4 is a block diagram showing an outline of the behavior analysis apparatus according to the present invention.
  • the behavior analysis apparatus analyzes the input acoustic information and generates acoustic analysis information representing the characteristics of the acoustic information (corresponding to the voice / acoustic analysis unit 10 shown in FIG. 1).
  • the time difference from the occurrence of the acoustic event specified by the acoustic analysis information to the occurrence of the event corresponding to the acoustic event in the input video captured by the crowd is determined, and the time difference information indicating the time difference is A behavior analysis unit that analyzes crowd behavior corresponding to an acoustic event using a generated time difference determination unit 2 (corresponding to the time difference determination unit 20 shown in FIG. 1), an input video, acoustic analysis information, and time difference information. 3 (corresponding to the crowd behavior analysis unit 30 shown in FIG. 1).
  • the time difference information may be information representing a distribution of time differences determined by the time difference determination unit.
  • the time difference is not a single value but is calculated as a distribution having a certain width, that is, the time difference (estimated value) determined by the time difference determination unit 2 varies. Even in this case, crowd behavior can be analyzed.
  • the behavior analysis unit 3 may analyze the crowd behavior based on the input video after the time corresponding to the time difference has elapsed since the occurrence of the acoustic event. According to such a configuration, it is possible to more reliably detect abnormal behavior of the crowd even when the time at which the abnormality is detected in the audio information is different from the time at which the abnormality is detected in the video information. it can.
  • the behavior analysis unit 3 calculates the probability that an event occurs based on the acoustic analysis information, and the event occurs based on the input video after the time indicated by the time difference has elapsed since the occurrence of the acoustic event.
  • the probability of occurrence may be calculated, and it may be determined whether an abnormality has occurred in the crowd behavior based on the calculated probabilities.
  • the abnormal state of the crowd can be determined from the result of integrating the acoustic analysis result and the video analysis result in consideration of the time difference. For example, it is possible to more accurately determine the abnormalities of the crowd behavior by determining whether there is an abnormality in the crowd behavior using a value obtained by integrating the calculated probabilities and a preset threshold value. Become.
  • the time difference determination unit 2 starts from the occurrence of the acoustic event based on the distance between the position where the acoustic information is acquired and the shooting area of the input video until the event corresponding to the acoustic event occurs in the input video. Such a time difference may be determined. According to such a configuration, the time difference between the acoustic event and the video event can be obtained according to the position of the microphone or camera.
  • the time difference determination unit 2 may calculate the time difference based on the distance between the position where the acoustic information is acquired and the shooting area of the input video, and the acoustic feature indicated by the acoustic analysis information. According to such a configuration, the time difference between the acoustic event and the video event can be obtained with higher accuracy based on the volume and type of sound indicated by the acoustic analysis information. That is, the time difference determination unit 2 can obtain a time difference in consideration of the volume and type of sound input by the microphone.
  • the behavior analysis unit 3 (corresponding to the crowd behavior analysis unit 30 shown in FIG. 3) classifies the type of event of the crowd behavior based on the acoustic analysis information, and generates event classification information indicating the classification result, Based on the event classification information, at least one of adjustment of parameters used for crowd behavior analysis and algorithm switching may be performed. According to such a configuration, it is possible to adjust or switch video crowd behavior analysis parameters and algorithms in accordance with the type of crowd behavior event. Thereby, crowd behavior can be analyzed more accurately. Further, it is possible to efficiently use calculation resources. Further, even when there are a plurality of cameras, it is possible to appropriately distribute calculation resources allocated to video analysis of each camera.
  • the behavior analysis unit 3 may calculate the likelihood representing the likelihood of a specific event as event classification information. According to such a configuration, even if it is not possible to classify the types of events of crowd behavior, it is possible to adjust or switch video crowd behavior analysis parameters and algorithms according to the crowd behavior events.
  • the behavior analysis unit 3 stores, in the crowd behavior determination result, an instruction for causing the predetermined device to execute a predetermined operation together with the analysis result of the crowd behavior, and outputs the crowd behavior determination result to the predetermined device. May be. According to such a configuration, for example, an alert can be reported to the security room. Further, the emergency exit can be opened by outputting the crowd action determination result to a device or the like that controls the emergency door. In addition, it is possible to output an image of a camera that is likely to show a criminal candidate person to the display device in the security room. In addition, the direction of the camera, the zoom rate, and the like can be controlled so that the face of the criminal candidate person can be easily photographed.
  • An acoustic analysis unit that analyzes input acoustic information and generates acoustic analysis information that represents characteristics of the acoustic information, and a crowd was photographed after an acoustic event specified by the acoustic analysis information occurred Determining a time difference required until an event corresponding to the acoustic event occurs in the input video, and generating a time difference information representing the time difference; the input video; the acoustic analysis information; and the time difference information. And a behavior analysis unit that analyzes a crowd behavior corresponding to the acoustic event.
  • time difference information is information representing a distribution of the time difference determined by the time difference determination unit.
  • the behavior analysis unit calculates the probability that an event will occur based on the acoustic analysis information, and the event occurs based on the input video after the time indicated by the time difference has elapsed since the occurrence of the acoustic event. 4.
  • the behavior analysis device according to any one of supplementary notes 1 to 3, which calculates a probability of occurrence of an anomaly and determines whether or not an abnormality has occurred in the crowd behavior based on the calculated probabilities.
  • the time difference determination unit calculates the time difference based on the distance between the position where the acoustic information is acquired and the shooting area of the input video, and the acoustic feature indicated by the acoustic analysis information.
  • the behavior analysis apparatus according to any one of the above.
  • the behavior analysis unit classifies the types of events of the crowd behavior based on the acoustic analysis information, generates event classification information indicating the classification result, and performs the crowd behavior analysis based on the event classification information.
  • the behavior analysis device according to any one of appendix 1 to appendix 6, which performs at least one of adjustment of parameters to be used and algorithm switching.
  • the behavior analysis unit stores, in the crowd behavior determination result, an instruction for causing the predetermined device to execute a predetermined operation together with the analysis result of the crowd behavior, and the crowd behavior determination result is stored in the predetermined device.
  • the behavior analysis device according to any one of supplementary notes 1 to 8 to be output.
  • the input acoustic information is analyzed, the acoustic analysis information showing the characteristic of the said acoustic information is produced
  • video Determine the time difference required until an event corresponding to the acoustic event occurs, generate time difference information representing the time difference, and respond to the acoustic event using the input video, the acoustic analysis information, and the time difference information
  • Supplementary note 15 Any one of Supplementary note 10 to Supplementary note 14 for calculating the time difference based on the distance between the position where the acoustic information is acquired and the shooting area of the input video, and the acoustic feature indicated by the acoustic analysis information The behavior analysis method described in one.
  • a behavior analysis program for executing a process of analyzing a crowd behavior corresponding to the acoustic event.
  • the time difference information is the behavior analysis program according to supplementary note 19, which is information indicating the distribution of the determined time difference.
  • the probability that an event will occur is calculated on the computer based on the acoustic analysis information, and the event occurs based on the input video after the time indicated by the time difference has elapsed since the occurrence of the acoustic event.
  • the behavior analysis program according to any one of supplementary note 19 to supplementary note 21, which calculates a probability and executes a process of determining whether or not an abnormality has occurred in the crowd behavior based on the calculated probabilities.
  • Supplementary note 19 to Supplementary note that cause the computer to execute a process of calculating a time difference based on the distance between the position where the acoustic information is acquired and the shooting area of the input video, and the acoustic feature indicated by the acoustic analysis information
  • the behavior analysis program according to any one of 23.
  • the computer stores an analysis result of the crowd behavior together with an instruction for causing the predetermined device to execute a predetermined operation in the crowd behavior determination result, and outputs the crowd behavior determination result to the predetermined device.
  • the behavior analysis program according to any one of supplementary note 19 to supplementary note 26 that causes the process to be executed.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Multimedia (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Acoustics & Sound (AREA)
  • Computational Linguistics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Alarm Systems (AREA)
  • Closed-Circuit Television Systems (AREA)
  • Image Analysis (AREA)

Abstract

L'invention porte sur un dispositif d'analyse d'action, comprenant: une unité d'analyse audio (1) qui analyse des informations audio introduites, et génère des informations d'analyse audio qui représentent une caractéristique des informations audio; une unité de détermination de différence temporelle (2) qui détermine une différence temporelle entre la survenue d'un événement audio qui est identifié par les informations d'analyse audio et la survenue d'un événement dans une vidéo introduite dans laquelle un groupe est filmé qui correspond à l'événement audio; et une unité d'analyse d'action (3) qui analyse une action de groupe correspondant à l'événement audio, à l'aide de la vidéo introduite, des informations d'analyse audio et de la différence de temps.
PCT/JP2014/001745 2013-04-26 2014-03-26 Dispositif d'analyse d'action, procede d'analyse d'action et programme d'analyse d'action WO2014174760A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2015513507A JP6344383B2 (ja) 2013-04-26 2014-03-26 行動解析装置、行動解析方法および行動解析プログラム
US14/786,931 US9761248B2 (en) 2013-04-26 2014-03-26 Action analysis device, action analysis method, and action analysis program

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2013093215 2013-04-26
JP2013-093215 2013-04-26

Publications (1)

Publication Number Publication Date
WO2014174760A1 true WO2014174760A1 (fr) 2014-10-30

Family

ID=51791350

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2014/001745 WO2014174760A1 (fr) 2013-04-26 2014-03-26 Dispositif d'analyse d'action, procede d'analyse d'action et programme d'analyse d'action

Country Status (3)

Country Link
US (1) US9761248B2 (fr)
JP (1) JP6344383B2 (fr)
WO (1) WO2014174760A1 (fr)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2017212680A (ja) * 2016-05-27 2017-11-30 キヤノン株式会社 画像処理装置、画像処理方法およびプログラム
JP2018092468A (ja) * 2016-12-06 2018-06-14 株式会社村田製作所 防犯監視システム及びその制御方法並びにコンピュータプログラム
WO2020256152A1 (fr) * 2019-06-21 2020-12-24 トニー シュウ Dispositif de traitement d'informations, procédé de traitement d'informations et programme
WO2022185569A1 (fr) * 2021-03-02 2022-09-09 株式会社日立製作所 Système d'analyse vidéo et procédé d'analyse vidéo
WO2023002563A1 (fr) * 2021-07-20 2023-01-26 日本電気株式会社 Dispositif de surveillance, système de surveillance, procédé de surveillance, et support non transitoire lisible par ordinateur dans lequel est stocké un programme
WO2023026437A1 (fr) * 2021-08-26 2023-03-02 日本電気株式会社 Dispositif de surveillance, système de surveillance, procédé de surveillance, et support non transitoire lisible par ordinateur dans lequel est stocké un programme

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018044553A1 (fr) * 2016-08-29 2018-03-08 Tyco Fire & Security Gmbh Système et procédé d'identification acoustique de coups de feu tirés à l'intérieur
CN108182950B (zh) * 2017-12-28 2021-05-28 重庆大学 改进的经验小波变换的公共场所异常声音特征分解与提取方法
TW201931863A (zh) * 2018-01-12 2019-08-01 圓剛科技股份有限公司 多媒體訊號的同步設備及其同步方法
CN111819530A (zh) * 2018-03-09 2020-10-23 三星电子株式会社 电子设备中用于增强用户体验的电子设备和设备上方法
US11100918B2 (en) * 2018-08-27 2021-08-24 American Family Mutual Insurance Company, S.I. Event sensing system
US11228791B1 (en) * 2018-10-25 2022-01-18 Amazon Technologies, Inc. Automatically processing inputs to generate content
CN111062337B (zh) * 2019-12-19 2023-08-04 北京迈格威科技有限公司 人流方向检测方法及装置、存储介质和电子设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002021441A1 (fr) * 2000-09-06 2002-03-14 Hitachi, Ltd. Detecteur de comportement anormal
JP2004139261A (ja) * 2002-10-16 2004-05-13 Matsushita Electric Ind Co Ltd 監視モニター装置
JP2007228459A (ja) * 2006-02-27 2007-09-06 Ikegami Tsushinki Co Ltd 監視システム
JP2010232888A (ja) * 2009-03-26 2010-10-14 Ikegami Tsushinki Co Ltd 監視装置
JP2013131153A (ja) * 2011-12-22 2013-07-04 Welsoc Co Ltd 自律型防犯警戒システム及び自律型防犯警戒方法

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5231492A (en) * 1989-03-16 1993-07-27 Fujitsu Limited Video and audio multiplex transmission system
JP2000217095A (ja) * 1999-01-20 2000-08-04 Mitsubishi Electric Corp 画像表示装置
US7194752B1 (en) * 1999-10-19 2007-03-20 Iceberg Industries, Llc Method and apparatus for automatically recognizing input audio and/or video streams
JP2002314987A (ja) * 2001-04-16 2002-10-25 Hitachi Ltd 監視システム
US8009966B2 (en) * 2002-11-01 2011-08-30 Synchro Arts Limited Methods and apparatus for use in sound replacement with automatic synchronization to images
EP1593272B1 (fr) * 2003-02-14 2017-08-30 Thomson Licensing DTV Synchronisation automatique de services media de contenus media fondes sur des fichiers audio et des fichier video
US20050219366A1 (en) * 2004-03-31 2005-10-06 Hollowbush Richard R Digital audio-video differential delay and channel analyzer
CA2563478A1 (fr) * 2004-04-16 2005-10-27 James A. Aman Systeme automatique permettant de filmer en video, de suivre un evenement et de generer un contenu
JP4506381B2 (ja) * 2004-09-27 2010-07-21 沖電気工業株式会社 単独行動者及びグループ行動者検知装置
US20090207277A1 (en) * 2008-02-20 2009-08-20 Kabushiki Kaisha Toshiba Video camera and time-lag correction method
US8342966B2 (en) * 2008-10-24 2013-01-01 Cfph, Llc Wager market creation and management
US20130212507A1 (en) * 2010-10-11 2013-08-15 Teachscape, Inc. Methods and systems for aligning items of evidence to an evaluation framework
IT1403658B1 (it) * 2011-01-28 2013-10-31 Universal Multimedia Access S R L Procedimento e mezzi per scandire e/o sincronizzare eventi audio/video
US8704904B2 (en) * 2011-12-23 2014-04-22 H4 Engineering, Inc. Portable system for high quality video recording
EP2763393A4 (fr) * 2012-01-06 2015-07-01 Asahi Chemical Ind Dispositif de formation d'images et dispositif de traitement des informations
KR101932535B1 (ko) * 2012-08-27 2018-12-27 한화테크윈 주식회사 실내 감시 시스템 및 실내 감시 방법

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2002021441A1 (fr) * 2000-09-06 2002-03-14 Hitachi, Ltd. Detecteur de comportement anormal
JP2004139261A (ja) * 2002-10-16 2004-05-13 Matsushita Electric Ind Co Ltd 監視モニター装置
JP2007228459A (ja) * 2006-02-27 2007-09-06 Ikegami Tsushinki Co Ltd 監視システム
JP2010232888A (ja) * 2009-03-26 2010-10-14 Ikegami Tsushinki Co Ltd 監視装置
JP2013131153A (ja) * 2011-12-22 2013-07-04 Welsoc Co Ltd 自律型防犯警戒システム及び自律型防犯警戒方法

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2017212680A (ja) * 2016-05-27 2017-11-30 キヤノン株式会社 画像処理装置、画像処理方法およびプログラム
JP2018092468A (ja) * 2016-12-06 2018-06-14 株式会社村田製作所 防犯監視システム及びその制御方法並びにコンピュータプログラム
WO2020256152A1 (fr) * 2019-06-21 2020-12-24 トニー シュウ Dispositif de traitement d'informations, procédé de traitement d'informations et programme
JP2021002229A (ja) * 2019-06-21 2021-01-07 株式会社Ｐｏｌｙｐｈｏｎｉｅ 情報処理装置、情報処理方法、及びプログラム
WO2022185569A1 (fr) * 2021-03-02 2022-09-09 株式会社日立製作所 Système d'analyse vidéo et procédé d'analyse vidéo
WO2023002563A1 (fr) * 2021-07-20 2023-01-26 日本電気株式会社 Dispositif de surveillance, système de surveillance, procédé de surveillance, et support non transitoire lisible par ordinateur dans lequel est stocké un programme
WO2023026437A1 (fr) * 2021-08-26 2023-03-02 日本電気株式会社 Dispositif de surveillance, système de surveillance, procédé de surveillance, et support non transitoire lisible par ordinateur dans lequel est stocké un programme

Also Published As

Publication number Publication date
US20160078883A1 (en) 2016-03-17
JP6344383B2 (ja) 2018-06-20
JPWO2014174760A1 (ja) 2017-02-23
US9761248B2 (en) 2017-09-12

Similar Documents

Publication Publication Date Title
JP6344383B2 (ja) 行動解析装置、行動解析方法および行動解析プログラム
JP5560397B2 (ja) 自律型防犯警戒システム及び自律型防犯警戒方法
JP6532106B2 (ja) 監視装置、監視方法および監視用プログラム
JP4861723B2 (ja) 監視システム
Andersson et al. Fusion of acoustic and optical sensor data for automatic fight detection in urban environments
JP6682222B2 (ja) 検知装置及びその制御方法、コンピュータプログラム
Conte et al. An ensemble of rejecting classifiers for anomaly detection of audio events
JP5047382B2 (ja) ビデオ監視時に移動物体を分類するシステムおよび方法
WO2011025460A1 (fr) Procédé et système de détection d'événement
WO2014174737A1 (fr) Dispositif de surveillance, méthode de surveillance et programme de surveillance
Choi et al. Selective background adaptation based abnormal acoustic event recognition for audio surveillance
KR102518615B1 (ko) 이상 음원을 판단하는 복합 감시 장치 및 방법
KR20130097490A (ko) 음향 정보 기반 상황 인식 장치 및 방법
Park et al. Sound learning–based event detection for acoustic surveillance sensors
Potharaju et al. Classification of ontological violence content detection through audio features and supervised learning
Varghese et al. Video anomaly detection in confined areas
KR101407952B1 (ko) 엘리베이터 방범시스템 및 방법
JP2007114885A (ja) 画像の類似性による分類方法及び装置
EP4171023A1 (fr) Procédé mis en uvre par ordinateur et programme informatique pour générer une vignette à partir d'un flux ou d'un fichier vidéo, et système de vidéosurveillance
JP4175180B2 (ja) 監視通報システム
JP4859130B2 (ja) 監視システム
Ntalampiras Audio surveillance
Spadini et al. Sound event recognition in a smart city surveillance context
Dedeoglu et al. Surveillance using both video and audio
Uzkent et al. Pitch-range based feature extraction for audio surveillance systems

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 14787662

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2015513507

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 14786931

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 14787662

Country of ref document: EP

Kind code of ref document: A1